xoL5zo1O86@OpenReview

Total: 1

#1 Reasoning Is Not a Race: When Stopping Early Beats Going Deeper [PDF] [Copy] [Kimi2] [REL]

Authors: Mohan Zhang, Jiaxuan Gao, Shusheng Xu, Yi Wu

We study the use of Process Reward Models (PRMs) for guiding Long Chain-of-Thought (CoT) reasoning in large language models. Although PRMs deliver fine-grained feedback in standard tasks, PRM-guided beam search does not consistently outperform PRM-free approaches in long CoT reasoning. We trace this shortfall to a "step quality degradation''—the expected step quality shows concave behavior, yielding unimodal or monotonically declining trends. To counteract this, we propose Z-Score Guided Early Stopping (ZGES), which halts search at the detected quality peak using local PRM-reward z-scores. Across multiple math benchmarks and model scales, ZGES outperforms both standard PRM-guided beam search and the PRM-free methods. Ablation studies further highlight the advantages and robustness of ZGES’s adaptive stopping mechanism.

Subject: NeurIPS.2025 - Poster